关键词: 光流, 块匹配, 并行加速

Abstract: Barnes approximate nearest neighbor algorithm is currently a well performance patch matching algorithm. This paper applies it in the compute of dense optical flow, and compares results with other two dense optical flow algorithms implemented by OpenCV. To make the Barnes algorithm easy to parallelize on GPU, the process of propagation is modified. Experimental results show that the parallel optical flow algorithm is more than two times faster than the original algorithm, the accuracy is nearly close to the original algorithm and better than the two dense optical flow algorithms implemented by OpenCV.

Key words: optical flow, patch match, parallelize